Automatic Melodic Grammar Generation for Polyphonic Music Using a Classifier System

In this paper, we propose a method to generate newmelodic styles (melodics) in the automatic composition of polyphonic music. In the proposed method, a melodic style is represented as a grammar that consists of rewriting rules, and the rewriting rules are generated by a classifier system, which is a genetics-based machine learning system. In the previous studies of grammatical approaches, the problem of how to treat polyphony and that of generating new melodic styles automatically haven’t been studied very intensively. Therefore, we have chosen to tackle those problems. We modeled the generative process of polyphonic music as asynchronous growth by applying rewriting rules in each voice separately. In addition, we developed a method to automatically generate grammar rules, which are the parameters of the polyphony model. The experimental results show that the proposed method can generate grammar rules and polyphonic music pieces that have characteristic melodic styles.